Development of an Integrative Model for Reconstructing Dynamic Transcriptional Regulatory Networks from High Throughput Data

The execution of transcriptional gene regulation mainly relies on the transcription factors (TFs, or the sequence-specific DNA-binding factors), which bind to specific DNA sequences or interact with other proteins, to promote or block the recruitment of RNA polymerase thereby controlling the transcription of the target genes. Due to the fact that gene transcription is essentially a nonlinear and dynamic process that is cooperatively regulated by a group of TFs, identification of the transcriptional regulatory model is a very challenging problem.

Integrative analysis on gene expression and TF-DNA binding profile or DNA motifs has been increasingly used to reconstruct both functionally and physically meaningful regulatory relationships. However, the current approaches are mainly devoted to reconstruct the ‘static’ networks, which cannot describe the dynamics in gene transcription. Besides, inferences of gene regulations are often based on the linear correlations between genes and regulators thereby cannot accommodate the nonlinear and interactive features of gene regulation.

In this project, we propose an integrative model to identify the transcriptional regulatory functions using high throughput data of gene expression and TF-DNA binding signal. Our model is developed based on the biologically sound cis-regulatory logics and the canonical transcriptional kinetic equations. Therefore, our model has clear biological meanings.

We first time showed that, whatever the underlying regulatory logic is, the dynamic gene expression can be uniformly decomposed into a linear function of the composite variables of the binding occupancies of a group of TFs, which is also called the model equation in our theory. Further, we showed that the coefficients in the model equation can be explicitly mapped backed to the original regulatory configuration (including both the logic and kinetic parameters) through a group of parameters functions.

During model identification, the coefficients in the model equation will be obtained by solving the constrained optimization problem using the dynamic expression data of the target gene as well as the binding occupancies of the candidate TFs. Consequently, the best configuration (both the logic and kinetic parameters) of TFs regulation will be identified from the coefficients using the pre-specified parameter functions.

Our theory will be validated using both simulation networks and real-world experimental data. In particular, we will reconstruct the condition-specific dynamic transcriptional regulatory networks of cell cycle control for normal and cancer cells. The derived networks will be validated using enrichment analysis, dynamic prediction analysis, functional analysis, as well as experiment validations. By comparing the regulatory patterns of the condition- specific TRNs, we could potentially unravel new molecular mechanisms of cell cycle desregulation in human cancer.

The main novelties of our approach include: 1) it first time convert the nonlinear regulatory function into a linear model equation, in which the coefficients can be explicitly mapped to a specific regulatory configuration including the regulatory logic and its kinetic parameters; 2) our model first time provides a natural way to integrate different types of experimental data to identify the regulatory model; 3) our approach can simultaneously identify both the regulatory logic and the kinetic parameters thereby maintaining their consistency in the developed networks.

Our method is limited in the sense that it only models the regulatory effects of the transcription factors. With the absence of some functional proteins, such as coactivators, chromatin remodelers, histone acetylases, deacetylases, kinases, and methylases, the derived networks may only assemble a partial loop of gene regulation.